The DONALD study is an ongoing, prospectively designed open cohort study conducted in Dortmund, Germany. Since 1985, data on diet, growth, and developmental and metabolic factors are collected continuously from infancy (age 3 months) to adulthood. Approximately 30–35 healthy infants from Dortmund and surrounding communities, whose mothers and/or fathers have a sufficient level of the German language, are recruited every year via personal contacts, maternity wards, or pediatric clinics. The examination schedule includes quarter-yearly examinations in infancy, half-yearly examinations in the second year of life, and annual examinations thereafter until young adulthood. Among others, examinations include anthropometric measurements, lifestyle questionnaires, and a 3-day food record. Chronotype assessment in the DONALD study started in 2014 for participants from 9 years onwards by use of the Munich Chronotype Questionnaire (MCTQ) [17, 18]. Detailed information regarding the study design can be found elsewhere [19].

Until July 2020, n = 652 study participants completed the MCTQ (N = 1.237 Q). Questionnaires collected during the first 2 weeks after the time change in Germany from standard winter to summer time or vice versa [10] (N = 95) and questionnaires with missing values (N = 38) were excluded. The sample was further reduced due to missing information on in parallel collected dietary data leaving a total number of n = 259 participants (N = 461 questionnaires). Of those, (n = 196 out of 172 families) were adolescent (aged 9–16 years) [20] and provided N = 401 questionnaires (Appendix Fig. 3).

Dietary intake in the DONALD study is assessed by use of 3-day weighed dietary records. The participants are free to choose the consecutive days, meaning that week and weekend days can be recorded. All foods and beverages consumed, as well as leftovers, are weighed and recorded over three consecutive days by the parents or by the older participants themselves with the use of regularly calibrated electronic food scales [initially Soehnle Digita 8000 (Leifheit AG, Nassau,Germany), now WEDO digi 2000 (Werner Dorsch GmbH, Muenster/Dieburg, Germany)]. When exact weighing is not possible, household measures (e.g., spoons and cups) are allowed for semi-quantitative recording. Information on recipes and on the types and brands of food items consumed is also requested. Additionally, participants record the time of every eating occasion. Energy and macronutrient intakes were calculated using the continuously updated in-house nutrient database LEBTAB [21], which is based on German standard food composition tables. Energy and nutrient contents of commercial food products are calculated by recipe simulation using labeled nutrient contents and ingredients. Macronutrients were considered as percentages of total energy intake (TEI). Subsequently, the individual means of TEI and macronutrient intakes were calculated from the three record days. The validity of dietary recording was previously evaluated by Bokhof et al. [22].

The MCTQ [17, 18] includes questions regarding sleep and wake times during the week and weekend. The individual chronotype (continuously in h:min) was calculated as the Midpoint of Sleep, i.e., the half-way point between sleep-onset and sleep-end on free days (MSF) [17]. If applicable, the MSF is corrected for “oversleep” on free days to account for sleep-debt accumulated over the week (MSFsc) [18].

Boys and girls are different in their individual level of lateness; girls tend to be earlier in comparison to boys [23]. To account for age and sex differences in chronotype, we derived MSFsc residuals for all observations independent of age and sex and ranked them by the group of two (PROC RANK in SAS®). Afterward, we calculated median MSFsc hours stratified by sex for adolescents with earlier and later chronotype (Fig. 1).

Diet-chrono alignment scoring (DCAS). MSFsc Midpoint of sleep (chronotype) corrected for sleep-debt accumulated over the workweek; h hours; m minutes. Alignment between chronotype and time of ‘highest caloric intake’ was expressed as the number of hours diverging from the chronotype-specific median eating time of ‘highest caloric intake’. Diet data were assessed by 3-day weighed dietary records, from which we derived average timing of ‘highest caloric intake’. Earlier eating resulted in positive divergence from the chronotype-specific reference eating time (defined as the median eating time of the specific chronotype), whereas later eating resulted in negative divergence from the reference for early chronotypes and vice versa for late chronotypes. MSFsc was regressed on age and sex. We ranked the residuals by the group of two and show in this figure the sex-specific MSFsc hours for early and late chronotypes

In a first step, we calculated the sum of energy intake per eating occasion, e.g., all foods and beverages consumed within a 30-min time period were summarized into one eating occasion; and all eating occasions < 10 kcal were added to the previous eating occasion [10]. From this, we derived the time of the day at which the greatest amount of energy (> 20% of total daily energy intake) was consumed (timing of ‘highest caloric intake’ per day). Then, these individual eating times were averaged and defined as ‘highest caloric intake’ derived from 3-day dietary records.

In a second step, age- and sex-adjusted median eating times, for which we applied the residual method, were derived for adolescents with earlier and later chronotype. The median eating time was defined based on the sample population and served as the reference point for the definition of diet-chrono-alignment. Third, individual diet-chrono-alignment score (DCAS) was calculated as the difference in hours between the chronotype-specific median eating time of ‘highest caloric intake’ and the individual eating time of ‘highest caloric intake’. Hence, earlier eating in comparison to the population median eating time of ‘highest caloric intake’ resulted in positive alignment for adolescents with an earlier chronotype measured in hours, whereas for adolescents with a later chronotype, later eating resulted in positive alignment (Fig. 1).

Adolescents were measured annually by trained nurses according to standard procedures, dressed in underwear and barefoot. Standing height is measured to the nearest 0.1 cm using a digital stadiometer. Weight was measured to the nearest 0.1 kg with an electronic scale (model 753 E; Seca, Hamburg, German). Skinfold thickness was measured on the right side of the body at the biceps, triceps, subscapular, and suprailiac sites to the nearest 0.1 mm with a Holtain caliper (Holtain Ltd., Crymych, UK) [19]. Sex- and age-independent body mass index (BMI kg/m2) standard deviation scores (SDS) were calculated using the German national reference data according to the LMS Method [24]. Percent body fat (%BF) was estimated from two skinfolds (triceps, subscapular) using age-specific Slaughter equations [25], for the subsequent calculation of FMI (fat mass/m2; where fat mass = body weight * BF%/100) and FFMI (fat free mass/m2; where fat free mass = body weight − fat mass). Since the distribution of FMI was skewed, log10-transformed values were used in the analyses.

Parents are interviewed regarding family characteristics (i.e., parental education, smoking, and persons in the household). Every 4 years, maternal body weight and height were measured with the same equipment as for the children on the child’s admission to the study center.

Adolescents’ pubertal status was defined in accordance with the onset and the end of pubertal growth spurt. Age at Take-Off (ATO) is the age of minimal growth velocity. Age at Peak Height Velocity (APHV) is the age of maximal height velocity. ATO and APHV were derived from the parametric Preece and Baines Model 1 [26], details are explained elsewhere [26, 27]. Under- and over-reporting of dietary intake was assigned if TEI was unrealistic in relation to the estimated basal metabolic rate (according to age- and sex-specific equations of Schofield [28]). Based on the pediatric cut-offs by Sichert-Hellert et al. [29] we detected 64 under-reporters (15.9%) and no over-reporting of dietary records.

All statistical analyses of the present evaluation were performed using SAS® procedures (version 9.4; Cary, NC, USA). The significance level was set at p < 0.05.

Linear mixed-effects regression models (PROC MIXED in SAS), including both fixed and random effects accounting for the nested nature of our data (children within families in the random statement) and the lack of independence between repeated observations on the same person. A further advantage is that the inclusion of all measurements is possible also in case of missing data for a specific point in time [30].

Repeated-measures regression models (PROC MIXED) were used to examine change-on-change associations of (a) the cross-sectional and (b) the longitudinal associations between the DCAS at first assessment and (Δ) body composition as well as (c) the respective changes in associations of (Δ) in DCAS and (Δ) BMI(-SDS), (Δ) FMI, and (Δ) FFMI over time [30, 31]. Variables of change were calculated by subtracting the baseline value from value at each year of assessment with 0 difference at first assessment.

Covariates for model adjustment were selected according to known predictors of BMI, body composition, and timing of energy intake [16, 32, 33]. From here, we identified minimally sufficient adjustment sets (msas) using a diagram (Appendix Fig. 4) representing the relationships among the identified variables [34]. Besides age at baseline and time between first and subsequent measurements (basic model) [31], sex, energy intake (g/d), physical activity (high vs. low), age at take-off and age at peak height velocity as puberty markers, smoking in the household, social jetlag, season of record (spring/summer/autumn/winter), number of questionnaires, and underreporting of dietary intake were important covariates.

Model building was driven by the log-likelihood criterion to define the final crude model. The Akaike Information Criterion (AIC) and the Bayesian Information were used to select the correlation structure best describing the correlated nature of the data. Tests for effect modification were performed by the inclusion of interaction terms between DCAS and MSFsc or sex.

To manage missing data, we undertook multiple imputations, using the MI procedure in SAS and explored the pattern of missingness. We generated an imputed database containing five imputed versions to predict missing values for ATO (n = 72, 18%) and physical activity (n = 67, 17%). Final models were tested regarding multicollinearity, heteroscedasticity, and normal distribution of residuals. Finally, we excluded participants who used an alarm clock during the weekend (n = 59, 15%) in a sensitivity analyses and assessed the influence of ‘eveningness in energy intake’ on body compositional measures. ‘Eveningness in energy intake’ was defined as percentage of energy intake in the evening (after 6 pm until 11:15 pm) − percentage of energy intake in the morning (before 11 am starting at 5:15 am) [9].